HCEO WORKING PAPER SERIES

Working Paper

The University of Chicago1126 E. 59th Street Box 107

Chicago IL 60637

www.hceconomics.org

You’ve got mail: A randomised fieldexperiment on tax evasion∗

Kristina M. Bott Alexander W. Cappelen Erik Ø. SørensenBertil Tungodden

April 25, 2017

Abstract

We report from a large-scale randomized field experiment conducted on aunique sample of more than 15 000 taxpayers in Norway, who were likely tohave misreported their foreign income. We find that the inclusion of a moralappeal or a sentence that increases the perceived probability of detection in aletter from the tax authorities almost doubled the average self-reported foreignincome. The moral letter mainly works on the intensive margin, while thedetection letter mainly works on the extensive margin. We also show that thedetection letter has large long-term effects on tax compliance.

JEL Classification Numbers: C93, D63, H26.Keywords: Taxation, tax evasion, field experiment.

∗We would like to thank Nadja Dwenger, Uri Gneezy, Espen Moen, David Laibson, CamilleLandais, John List, Ragan Petrie, Agnar Sandmo, Laszlo Sandor, Emmanuel Saez, Paul Smeets, andKjetil Storesletten for extremely valuable comments and suggestions; Kenneth Andreassen, IngvildHoller Deisz, Marta Johanne Gjengedal, Anita Hallenstvedt, Pia Høst, Elin Imsland, Paul-GunnarLarsen, Dag Løvas, Bente Lundekvam, Sissel Madsen, Gitte Martensen, Vibeke Vik Nordang, ElinAam Svendsen, Irene Søreide, Thomas Tangen, Christine Osen Tefre, Marcus Zackrisson, and othersat Norwegian Tax Authority for their assistance in implementing the experiment; and Trine Sivert-sen Sommerlade at Bergen Brannvesen and Fredrik Naumann for pictures used in the experimentalmaterials. All authors: The Choice Lab, NHH Norwegian School of Economics. The project wasfinanced by support from the Research Council of Norway, research grant 236995 and the NOR-FACE Welfare State Future Program, and it was administered by The Choice Lab, NHH NorwegianSchool of Economics.

1 IntroductionA key challenge in all modern societies is to limit tax evasion, which causes largelosses in government revenues and create significant unfairness in society. It hasfor example been argued that the loss of government revenue amounts to 500 bil-lion USD in the US, corresponding to the size of the government deficit, and 11billion Euros in Greece, corresponding to 30% of the government deficit (Cebulaand Feige, 2012; Artavanis, Morse, and Tsoutsoura, 2015). Tax evasion is partic-ularly difficult to handle when the tax administration has to rely on self-reporteddata, since taxpayers have an economic incentive to underreport income (Alling-ham and Sandmo, 1972; Slemrod and Yitzhaki, 2002; Sandmo, 2005). The classicalapproach to increasing tax compliance has therefore been to reduce the economicincentives for tax evasion, by increasing the detection probability and penalties andcollecting more third-party information. In recent years, however, there has beena growing interest in understanding the extent to which moral motivation or morebroadly “tax morale” can play a role in increasing tax compliance in society (Slem-rod, 2007; Luttmer and Singhal, 2014).

To study the drivers of tax compliance, and in particular the role of moral mo-tivation, we conducted a large-scale field experiment together with the Norwegiantax administration on a unique sample of more than 15 000 taxpayers. The sampleconsisted of taxpayers who were likely to have misreported their foreign income inthe previous tax year, but who were not aware of the fact that the Norwegian taxadministration had information about this misreporting. Information about foreignincome is not included in the pre-populated tax return in Norway and the taxpay-ers therefore have to self-report this information. Historically, it has been difficultfor the tax authorities to verify the self-reported information because they have nothad access to third-party reports from foreign countries, but this has changed due torecent international collaboration among tax authorities.

The field intervention consisted of an information letter sent by the tax adminis-tration shortly before the taxpayers were to submit their tax return for the previousyear, where we randomly assigned taxpayers to receive different versions of a baseletter or to a control group that did not receive any letter. The base letter containedinformation about why and how to report foreign income income and the effect ofthis letter sheds light on whether the underreporting was driven by a lack of infor-mation about tax procedures. The main focus of this study, however, is to identifythe causal effects of introducing moral suasion and increasing the perceived de-tection probability on tax evasion, and thus we manipulate the base letter alongeach of these two dimensions in additional treatments. We study two versions ofmoral suasion, a fairness argument and a societal benefits argument for correctly re-porting foreign income. To investigate the importance of the detection probability

2

for tax evasion, we added information that we believed would make the taxpayersincrease their subjective probability of being audited. We study the effect of ourtreatment manipulations on self-reported foreign income in the following tax return(the follow-up year) and one year later (long term), and also provide additional ev-idence allowing us to cleanly identify the underlying mechanisms of the treatmenteffects.

Our main result is that moral suasion has a large and significant effect on self-reported foreign income. As shown in panel A in Figure 1, in the follow-up year,the average self-reported foreign income by the taxpayers who receive one of themoral letters is almost double the amount self-reported by those who receive thebase letter. There is no statistically significant difference in the effect of the twoversions of the moral letter. We also find a large effect of the detection letter, butthe moral letters and the detection letter affect different margins of the taxpayerbehavior. As shown in panel B of Figure 1, the detection letter has a large effecton the extensive margin: it increases the share of taxpayers reporting a positiveamount from 20% among those who receive the base letter to 33% among thosewho receive the detection letter. In contrast, the moral letters have a minor effecton the extensive margin, but a large effect on the intensive margin: the moral letterssignificantly increase the self-reported foreign income among the taxpayers whoalready report some foreign income. Further, we show that the base letter itselfhas some effect on self-reported foreign income compared to the group that didnot receive any letter, but overall, our study suggests that the underreporting is notprimarily driven by a lack of knowledge about how to report foreign income.

[ Figure 1 about here. ]

Our findings are robust across different subgroups (age, gender, Norwegian cit-izenship, socioeconomic status). For all subgroups, we observe that the moral let-ters and the detection letter increase the level of self-reported foreign income in thefollow-up year, where the moral letters typically work on the intensive margin andthe detection letter has a strong effect on the extensive margin. Finally, we providea set of results on the long-term effects of the intervention, where the main insightis that the detection letter has a large effect on the extensive margin even one yearafter the taxpayer received the letter, while there are no statistically significant long-term effects of the moral letters. These long-term findings suggest that the moralletters mainly worked through making the moral arguments salient when the tax-payer received the letter, while the detection letter caused the taxpayers to updateand sustain their beliefs about the detection probability.

Our paper contributes to the growing literature that uses field interventions tostudy tax compliance (Coleman, 1996; Blumenthal, Christian, and Slemrod, 2001;

3

Coleman, 2007; Kleven, Knudsen, Kreiner, Pedersen, and Saez, 2011; Ariel, 2012;Del Carpio, 2013; Fellner, Sausgruber, and Traxler, 2013; Castro and Scartascini,2015; Fellner et al., 2013; Hallsworth, 2014; Hallsworth, List, Metcalf, and Vlaev,2017; Dwenger, Kleven, Rasul, and Rincke, 2016).1 Evidence from these experi-ments on the drivers of tax evasion has been mixed and, in particular, most of thestudies have not been able to document that moral suasion may play an importantrole in reducing tax evasion. We believe that our study has three main strengthsthat may contribute to explain why we can cleanly identify strong effects of bothmoral suasion and an increase in the detection probability. First, we consider a sam-ple and situation where there is no third-party reporting and all taxpayers have anopportunity to evade taxes, while some of the previous studies have suffered froma significant part of the sample being restricted by third-party reporting (Klevenet al., 2011). Second, we carefully timed the distribution of the letters such that thetaxpayers received them close to the deadline for submitting the tax return, whilesome of the previous studies have had a significant lag between the field interventionand the moment of decision-making (Blumenthal et al., 2001; Fellner et al., 2013).Third, our experimental design allows for a clean test of whether it is moral sua-sion or an increase in detection probability that is driving the change in taxpayers’behavior. We compare the effect of the moral letters and the detection letter to theeffect of a base letter that only differs along the relevant dimension and we provideadditional survey evidence showing that the letters worked as intended. Overall,we therefore believe that our study provides novel, clean and robust evidence ofhow both moral suasion and an increase in detection probability may contribute toreduce tax evasion. Further, to our knowledge, we provide the first set of field evi-dence showing that moral suasion and detection probability affect different marginsof taxpayer behavior: moral suasion largely affecting the intensive margin, whiledetection probability largely affecting the extensive margin. Finally, we also add tothe existing literature by presenting long-term results on the effects of the field inter-vention, where we show that only the effect of the Detection treatment is sustainedin the long run.

Our results also speak to the growing literature in behavioural economics study-ing the role of moral motivation. These studies, mostly relying on lab experiments,have documented that moral motivation matters for people when making economicdecision (Fehr and Schmidt, 1999; Bolton and Ockenfels, 2000; Konow, 2000;Andreoni and Miller, 2002; Charness and Rabin, 2002; Engelmann and Strobel,2004; Cappelen, Drange Hole, Sørensen, and Tungodden, 2007; Cappelen, Moene,

1There are also interesting studies of tax compliance in the lab (Alm, McClelland, and Schulze,1992; Bo and Bo, 2009; Lamberton, Neve, and Norton, 2014) and by the use of observational data(Fisman and Wei, 2004; Rincke and Traxler, 2010; Carrillo, Emran, and Rivadeneira, 2012; Casaburiand Troiano, 2016).

4

Sørensen, and Tungodden, 2013; Fehr, Glatzle-Rutzler, and Sutter, 2013). More-over, in a related literature it has been shown that people do not always lie evenwhen they have an opportunity to do so and can gain from it (Gneezy, 2005; Eratand Gneezy, 2012). These lab experimental findings have sometimes been chal-lenged, because they are established in an artificial setting with small stakes (Levittand List, 2007). We demonstrate that moral motivation not only matters in the lab,but also in field settings involving large stakes: we find that the taxpayers who re-ceive the moral letters on average self-report 1 300 USD more in foreign incomethan individuals who receive the base letter.

The structure of the paper is as follows: Section 2 presents the setting for thefield experiment and the sample, while Section 3 provides details of the experimen-tal design. In Section 4, we provide a simple theoretical framework to guide ouranalysis, while we outline the empirical strategy in Section 5. Section 6 discussesthe results, while Section 7 provides some concluding remarks.

2 Background and sampleIn this section we first present how taxes are reported in Norway. We then describethe sample of taxpayers in our study.

2.1 Tax reporting in NorwayEvery year in April, the Norwegian tax administration (NTA) sends pre-populatedtax returns for the previous fiscal year to all Norwegian tax residents. The pre-populated tax return constitutes a preliminary tax statement and the taxpayer isrequired to add any missing information and correct potential mistakes before theend of April. If the taxpayer believes the information in the pre-populated tax returnto be correct and complete, he or she is not required to make any changes.

When filing their taxes, taxpayers are reminded to declare all income, both do-mestic and foreign, earned in the previous fiscal year. The domestic income is typi-cally for the most part included in the pre-populated tax form, based on third-partyreporting in Norway, but information about foreign income must be self-reportedby the tax subjects. Historically, it has been difficult for the NTA to control if tax-payers correctly report foreign income because there has been limited exchange ofinformation across national tax jurisdictions. Over the last few years, however, taxadministrations in a number of countries have increasingly provided informationabout the income and wealth that tax residents of other countries earn or hold intheir countries. As part of this development, the NTA has in recent years receivedreports from other tax administrations about Norwegian tax residents’ income and

5

wealth in the respective countries. These reports are referred to as Automatic Coun-try reports from Abroad (ACA; in Norwegian: Automatiske Kontrolloppgaver Ut-land). The exchange of such reports is a result of bilateral negotiations betweennational tax authorities, and not all countries exchange this type of informationwith the NTA.

The NTA received these reports with delay in the period we studied and thuscould not include information about foreign income in the pre-populated tax returns.However, the NTA could compare the self-reported foreign income in the domestictax returns with the foreign income recorded in the ACA-report at a later date, andthis comparison forms the basis for our study.

2.2 The sampleFor the fiscal year 2011, which is the baseline year of this study, the NTA receivedACA-reports for around 40 000 Norwegian tax residents. The NTA estimated that17 899 of these had self-reported between 2 000 NOK and 200 000 NOK (equiva-lent to approximately 350 – 35 000 USD in 2011) less in foreign income than statedin the ACA-reports for the income year 2011, and this group was the point of de-parture of the present study.2 These individuals were not aware of the fact that theNTA had information about their incorrect reporting of foreign income, and the taxauthorities did not act on this information until after the taxpayers had self-reportedforeign income for 2012, which is the follow-up year in this study. A small subsetof the group that self-reported incorrectly was randomly selected to be part of apractical policy experiment that focused on a specific applied question of interestfor the tax authorities, while the rest, 15 708 individuals, constitute the sample forthe present paper.3

Table 1 and Table 2 provide two sets of comparisons for the baseline year; acomparison between the general population and the tax subjects with an ACA reportabout foreign income and a comparison of those who self-reported incorrectly andthose who self-reported correctly foreign income. Tax subjects are classified asindividuals who self-reporter correctly if they have misreported less than 2000 NOKin the baseline year. From the left part of Table 1, we observe that compared to thegeneral population, the tax subjects with foreign income are more likely to be non-Norwegian citizens, a large share of whom are from other Nordic countries. We alsoobserve that the individuals with foreign income are slightly more likely to be maleand self-employed and are on average a few years older than the general population.

2We had to exclude 137 individuals, for whom the NTA only had incomplete ACA-reports.3The practical policy experiment tested the usefulness of a weblink providing further information

about how to report foreign income.

6

From the right part of Table 1, we observe that those self-reported incorrectly andthose who self-reported correctly are very similar on the background characteristics,with the exception of those who self-reported incorrectly on average being olderthan those who self-reported correctly.

[ Table 1 about here. ]

Table 2 provides descriptive statistics on income, wealth, and misreporting forthe baseline year. From panels A and B, we observe that the tax subjects withACA-reports have more income and wealth than the general population. Thosewho self-reported incorrectly have lower income than the those who self-reportedcorrectly, while the two groups have the same level of wealth. Those who self-reported incorrectly have the same level of income as the general population inNorway, but significantly more wealth.

[ Table 2 about here. ]

From panels C and D, we observe that average foreign income in the ACA-reports is 44 902 NOK and the mean amount misreported is 8 866 NOK. Those whoself-reported incorrectly have significantly more foreign income in the ACA-reportsthan those who self-reported correctly, 56 280 NOK versus 36 852 NOK, and self-report only 51 percent of it to the tax authorities. Those who self-reported correctlyreport 5 049 more than what is stated in the reports from the tax administrations inother countries, which might reflect that the ACA-reports do not capture all foreignincome for the tax subjects (both because the ACA-report from each country islikely to be incomplete and because the NTA does not receive information from alltax authorities across the world).

3 Experimental designThe basic structure of the experimental design is that all individuals in our samplereceived the pre-populated tax returns for the follow-up year in week 14 of 2013,and individuals in the treatment arms then received a letter from the Norwegian taxauthorities in week 15 about how to handle foreign income in the tax return; see acopy of the base letter in Figure A1 in the appendix.4

4Complete translations of all the letters are provided in Appendix B. Our experiment has anintention-to-treat design, since we do not know how many of the taxpayers actually read the letter.The likelihood of reading the letter, however, should not differ across treatments, since there wereno treatment differences in the design of the envelopes. To test whether tax subjects read a letter

7

The individuals could make changes to the pre-populated tax return, includingself-reporting of foreign income, until the deadline in week 18. The individualswere randomly allocated into a control group (No letter) or one of three treatmentarms (Base treatment, Moral treatments, Detection treatment).5 The individuals inthe control group did not receive any letter from the tax authorities, while the indi-viduals in the different treatment arms received different versions of the base letter.Our main interest is whether receiving such a letter increased the self-reported for-eign income for the follow-up year. We also have long-term data that allow us tostudy whether the letter intervention affected the self-reported foreign income oneyear later.

3.1 Base treatmentSince any letter from the tax authorities may cause a change in behavior for a num-ber of reasons (fear of detection, moral salience, or better knowledge about howto proceed with the reporting), we included a treatment where the tax residentsreceived a letter that only contained general information about how to self-reportforeign income (Base treatment).6 The letter consists of three paragraphs, the firstexplaining why the reader receives this letter. It refers to the fact that the Norwe-gian economy is becoming more international, and that an increasing number oftaxpayers has income from abroad. The taxpayer is then told that the NTA wouldlike to inform him or her about how this type of income is taxed and how it shouldbe reported.

The second paragraph of the letter states that all Norwegian tax residents areliable to pay taxes to Norway on all income and assets, even on foreign incomeand foreign assets unless otherwise specified in Norway’s tax treaties with other

from the tax authorities, the NTA conducted an independent survey where they sent out a versionof the base letter to 100 randomly selected taxpayers not taking part in this study. At the end of thefirst paragraph of this letter, the NTA told the taxpayer that they had been selected to test whetherindividuals actually read letters from the NTA, and therefore were asked to confirm that they had readthe letter by sending an sms to the NTA with the code provided in the letter. 29% of the recipientsof the letter confirmed that they had received it. The NTA then attempted to call the individuals whohad not responded and managed to get in touch with 63% of them (with the restriction that theymade a maximum of six attempts): 37% confirmed that they had read the letter, 3% confirmed thatthey had received it, but not read it, 19% stated that they had not received it, and 4% stated that theywere not sure whether they had received it.

5As shown in Table A1 in Appendix A, the treatments are balanced on the background variables.6We had two different versions of the base letter, one using active language, thus, addressing the

reader as “you”, and another using passive language (Bryan, Adams, and Monin, 2012). We do notfind an economically or statistically significant difference between these two versions of the baseletter on the amount self-reported (p = 0.775) or the share of individuals self-reporting a positiveamount (p = 0.884), and thus do not differentiate between them in the following analysis.

8

countries. This paragraph also provides a link to the website of the Norwegiantax administration and a phone number to a call centre in the tax administrationestablished for the purpose of this study.7 The final paragraph informs about howto proceed after having received the pre-populated tax returns, and adds a web-linkproviding further information about how to file the Norwegian tax return.

The Base treatment allows us to study whether lack of information about howto report foreign income is a main driver of the observed underreporting of foreignincome. It is, however, important to note that the treatment difference between theBase treatment and the No letter group provides an upper bound estimate of the roleof information, since the base letter may also trigger other mechanisms among thetaxpayers (including fear of detection and moral salience).

3.2 Moral and detection treatmentsTo identify as cleanly as possible the causal effects of introducing moral suasion andincreasing the perceived detection probability, we manipulated the base letter alongeach of these two dimensions in additional treatments. The treatment manipulationsonly introduced minor changes in the first paragraph of the letter; the rest of theletter was identical to the base letter.

We studied two types of moral appeals. In the Fairness treatment, the letterintroduced a fairness argument for reporting foreign income correctly, by includinga sentence that reminded the taxpayers of the fact that most Norwegians reportthe income earned in Norway correctly.8 Specifically, the following two sentenceswere added to the end of the first paragraph: “The great majority report informationabout their income and assets in Norway correctly and completely. In order to treatall taxpayers fairly, it is therefore important that foreign income and foreign assetsare reported in the same manner.”

In the Societal Benefits treatment, the letter introduced a societal benefits argu-ment for reporting foreign income correctly, by including a sentence that remindedthe taxpayers about the benefits to society resulting from taxation: “Your tax pay-

7In order to standardize the answers to the callers, the NTA provided the phone operators witha script of potential questions and answers. The phone operators were not aware of the call centrefacilitating a field experiment, they only knew that the authorities had sent out different letters todifferent individuals. In Table A2 in Appendix A, we provide an overview of the activity at the callcenter. 5% of the individuals receiving a letter approached the call centre, mainly asking questionsabout why they had received the letter and how to report foreign income. Significantly more individ-uals in the Detection treatment used the call centre than in the Base treatment (13.1% versus 3.6%,p < 0.001), while we see no difference between the Moral treatments and the Base treatment (3.7%called in the Moral treatments, p = 0.710).

8Almost all income earned in Norway is third-party reported to the tax authorities, as in the otherScandinavian countries; see also Kleven et al. (2011); Kleven (2014).

9

ment contributes to the funding of publicly financed services in education, healthand other important sectors of society”.9 In two additional treatments, we visualisedthe societal benefits from taxation by adding an attachment illustrating publicly fi-nanced services in health, education, infrastructure, and research; see Figure B6 inAppendix A. In one treatment, the attachment was combined with the base letter, inanother treatment it was combined with the societal benefits letter.

In the Detection treatment, the letter aimed to increase the perceived detectionprobability of the tax subject. We replaced the first sentence in the base letter: “TheNorwegian economy is becoming more internationalised, and an increasing numberof Norwegian taxpayers receive income and have assets abroad” with the sentence:“The tax administration has received information that you have had income and/orassets abroad in previous years”. The basic idea behind this treatment manipulationwas that providing information about the tax authorities’ knowledge about the indi-vidual activities abroad in previous years would make the tax subjects update theirsubjective beliefs about the likelihood of being audited in the coming fiscal year.

Table 3 provides an overview of the different treatment arms in the experiment.The experimental design allows for the following two main comparisons to studythe drivers of the misreporting of foreign income:

• The role of moral motivation: The comparison between the Moral treat-ments and the Base treatment identifies the causal effect of moral suasion ontaxpayer behavior.

• The role of the detection probability: The comparison between the Detec-tion treatment and the Base treatment identifies the causal effect of increasingperceived detection probability on taxpayer behavior.

When interpreting these treatment comparisons, we make two assumptions.First, we assume that the moral letters only manipulate the moral dimension rel-ative to the base letter; second, we assume that the detection letter only manipulatesthe perceived detection probability relative to the base letter. We tested these as-sumptions in an independent survey, where, as shown in Table A4 in Appendix A,we find support for the letters working as intended. Importantly, we find no evi-dence of the moral letters increasing the perceived detection probability among therecipients of the letters.

9This sentence may trigger a reciprocity motive for tax compliance, where individuals becomemore willing to pay taxes because they recognize it as an exchange for benefits that the state provides(Fehr and Gachter, 1998; Luttmer and Singhal, 2014).

10

4 Theoretical frameworkWe here provide a simple model of taxpayer behavior to guide our analysis and theinterpretation of the results, building on Cappelen et al. (2007); Sandmo (2012) .

Assume that the taxpayer has (only) foreign income y and self-reports r. Thetax on foreign income is t and the penalty on misreported income, if detected, is τ .After tax income is Y = y− tr if not detected and Z = y− tr− τ(y− r) if detected.Let us assume that the taxpayer has the following expected utility function:

EU(r; ·) = pu(Z)+(1− p)u(Y )−β (y− r)2, (1)

where p is the subjective probability of being detected and β ≥ 0 is the weightattached to the moral cost of misreporting. It follows straightforwardly that theinterior solution for the taxpayer is given by:

r = y− ∆us(r, t,τ, p)2β

, (2)

where ∆us(r, t,τ, p) = (τ − t)(1− p)u′(Y )− ptu′(Z). The second term in the firstorder condition captures the trade-off that determines the level of misreporting. Thenominator represents the costs in selfish terms of deviating from what would havebeen reported if β = 0 (since the optimal solution for such an individual wouldbe characterized by ∆us(r, t,τ, p) = 0), while the denominator shows the relativeimportance assigned to selfish versus moral costs. In the case where the taxpayeronly cares about the moral costs (β → ∞), the optimal choice is to self-report theforeign income correctly. More generally, the extent of underreporting will dependon the tax parameters, the shape of the utility function, the subjective detectionprobability, and the importance assigned to the moral cost of misreporting.

The Moral treatments and the Detection treatment aim to increase the weightattached to the moral costs of misreporting (β ) and to the subjective detection prob-ability (p), respectively, and thereby to increase the self-reported foreign income(r). We do not expect the moral letters permanently to change the moral motiva-tion of the taxpayers, but they may increase the weight attached to the moral costof misreporting by making the moral argument more salient. The Detection treat-ment provides new information to the taxpayers which should make them updatetheir subjective beliefs about the likelihood of being detected. Both treatments maywork on both the intensive and the extensive margin: The treatments may causean increase in the self-reported foreign income among the taxpayers who are at aninterior solution (the intensive margin) and an increase the share of individuals whoactually report some foreign income (the extensive margin).

11

5 Data and empirical strategyWe here provide an overview of the data and the empirical strategy for the mainanalysis and the heterogeneity analysis.

5.1 DataThe analysis uses data from the administrative records of the NTA. The main out-come variable of interest is self-reported foreign income in the tax return for 2012(follow-up year) and 2013 (long term). Further, we use the ACA-reports submittedto the NTA from 17 countries to calculate an estimate of the total foreign incomefor each individual.10 The administrative records also include data on age, gender,Norwegian citizenship, income, and wealth.

5.2 Empirical strategyIn the analysis, our main regression specification is:

yi,t = α +∑l∈L

βldil +δyi,b + γxi + εi, (3)

where yi,t is self-reported foreign income for individual i in year t. We let l indexa treatment in the set of treatments L, where dil is an indicator variable for whetherindividual i is in treatment l; yi,b is the self-reported foreign income in the baselineyear, and xi is a vector of background variables (including age, gender, Norwegiancitizenship, and a measure of socio-economic status defined by income and wealth).We report the specification where the Base treatment is the omitted category. Theestimated causal effect of treatment l relative to the Base treatment is then given bythe estimated βl coefficient.

We report both regressions where we pool all the Moral treatments, which pro-vides us with an estimate of the average causal effect of the moral treatments relativeto the baseline, and regressions where we estimate separately the causal effects ofthe fairness letter and the social benefits letter. Further, we report regressions wherewe pool all the treatment arms and define receiving a letter as the omitted category,which provides us with an estimate of the average causal effect of receiving any oneof the letters from the tax authorities. For all specifications, we report estimates forregressions both with and without the background variables.

10According to our agreement with the NTA, we are not allowed to list the countries providingACA-reports to the NTA.

12

The average causal effect is a combination of the effect on the extensive margin(causing people who would not have reported any foreign income without the let-ter to report some amount) and the effect on the intensive margin (causing peoplewho would have reported some foreign income without the letter to report more).We study the effect on the extensive margin by the same regression as in equation3, where the dependent variable is an indicator variable for self-reported foreignincome in year t being strictly positive. To shed some light on the effect on theintensive margin, we also report the effect on self-reported foreign income condi-tional on it being positive, but it is important to keep in mind that this conditionalvariable is determined both by the effect on the intensive margin and the effect onthe share of taxpayers who actually self-report positive foreign income.

To study whether there are large differences in how the treatments affect the par-ticipants, we also conduct an heterogeneity analysis using the background variablesage, gender, Norwegian citizenship, and socio-economic status. In this analysis, wetake the regression specification where we have pooled the Moral treatments as thepoint of departure. For each background variable, we partition the set of partici-pants I into G and I \G, with gi as an indicator variable for whether individual i isa member of G. To illustrate, if the relevant background variable is age, then wepartition the set of participants into two subsets, those who are below and above 60years. The indicator variable would then take the value one if the taxpayer is abovethose who are 60 years. In each case, we interact the indicator variable with thetreatment indicator dil ,

yi,t = α +βdil +θgidil +λgi +δyi,b + γxi + εi. (4)

The estimation sample is the participants in the l treatment and in the Base treat-ment. With this specification, the estimated treatment effect of being in treatment lfor individuals in the group G is β +θ , while it is β for individuals in group I \G.Now θ is the estimated difference in treatment effect between the two groups, andit provides the basis for a statistical test of whether the estimated heterogeneity isstatistically significant. The level effect on the self-reporting behavior of belongingto the group G is λ .

6 ResultsIn this section, we start out by examining how the treatments affected average self-reporting behavior in the follow-up year, before turning to an heterogeneity analysisof the treatments effects. In the final part, we report on long-term effects of theintervention.

13

6.1 Main analysisWe find significant underreporting of foreign income also in the follow-up year. Inour sample, 78 percent of the taxpayers had foreign income recorded in the ACA-reports for the follow-up year, on average 42 796 NOK, while only 20.7 percentof them self-reported to have foreign income, on average 15 485 NOK. We nowstudy whether the different letters caused the taxpayers to self-report more of theirforeign income.

[ Table 4 about here. ]

In Table 4, we report regressions on self-reported foreign income in the follow-up year based on equation 3. Columns (1)-(2) report estimates of the effect of notreceiving a letter, where all taxpayers who did receive a letter are pooled togetherand serve as the reference category. We observe from column (1) that receiving aletter has a large and highly statistically significant effect on self-reported foreignincome: it increases by 8875 NOK, more than 100%, from 8155 NOK in the Noletter group to 17 030 NOK in the treatment groups combined. As shown in column(2), the finding is robust to the inclusion of background variables on the amount ofself-reported foreign income in the baseline year, the amount of foreign incomerecorded in the ACA-reports, and personal and socio-economic characteristics ofthe taxpayer (p < 0.001, column (2)).11 Thus, we can state our first main result:

Result 1: A letter from the tax authorities has a large and statistically signifi-cant effect on the amount of self-reported foreign income.

In columns (3) – (4), we report estimated treatment effects for each of the lettersseparately, where the base letter now serves as the reference treatment. We observethat the No letter group reports a lower foreign income than the Base treatmentgroup. The estimated effect of the base letter is borderline significant (p = 0.113,column (4)), which suggests that the underreporting is partly driven by a lack ofinformation about how to report foreign income.

The estimates for the moral letters and the detection letter identify the causal ef-fects of adding moral suasion and increasing the detection probability. We observethat the effects are large and highly significant for all the three main treatments

11In panel A in Figure A3 in Appendix A, we show that there is no corresponding increase inthe requests for deductions in taxes based on taxes paid abroad; there is no statistically significantdifferences across treatments in the deduction amount requested (p = 0.551).

14

(p = 0.010 (Fairness), p = 0.034 (Societal Benefits), p = 0.028 (Detection), col-umn (4)).12 Moral suasion, in terms of a fairness argument or of making salientsocietal benefits from taxation, has a strong positive effect on self-reported foreignincome of almost the same magnitude as the introduction of information that in-creases the detection probability. As shown in columns (5) and (6), where we poolthe Moral treatments, moral suasion on average causes an increase in self-reportedforeign income of almost 70% (p = 0.008, column (6)), while the Detection treat-ment increases average self-reported foreign income by 80% (p = 0.028; column(6)).

Result 2: Including moral suasion or information that increases the detectionprobability in the letter from the tax authorities has an economically and statisti-cally highly significant effect on the amount of self-reported foreign income.

In Table 5, we study how the letters affect the extensive margin, i.e., the shareof taxpayers who reports to have positive foreign income in the follow-up year. Incolumns (1)-(2), we observe that receiving a letter from the tax authorities has alarge effect on the extensive margin. The share of taxpayers that report a positiveforeign income increases from 11% in the No letter group to 22% in the treatmentgroups combined, and, as shown in column (2), the effect is robust to the inclusionof the set of background variables (p < 0.001, column (2)).

[ Table 5 about here. ]

The treatments, however, affect the extensive margin very differently, as shownin columns (3)-(6). The base letter and the detection letter have a large and sta-tistically significant effect on the extensive margin (p < 0.001 (Base), p < 0.001(Detection), column (6)), increasing the share of individuals who self-report foreignincome from 11.4% to 20% and 33%, respectively. In contrast, the Moral treatmentsdo not on average have a significant effect on the extensive margin (p = 0.122, col-umn (6)). There is a small effect of the fairness letter (p = 0.007, column (4)),significantly weaker than that of the detection letter, while there is no effect on theextensive margin of making societal benefits salient (p = 0.478, column (4)).13

12In Table A3 in Appendix A, we report estimates for the different versions of the Societal Benefitstreatments (verbal, visual, verbal and visual), where we in all cases observe an increase in self-reported foreign income. We cannot reject that the three versions have the same effect on taxpayerbehavior (p = 0.59, column 2 in Table A3).

13In panel C in Figure A3 in Appendix A, we show that these results are robust to only consideringthe effect on the taxpayers that self-report some foreign income but do not request any deductions.

15

Result 3: The moral letters and the detection letter have very different effects onthe extensive margin. The detection letter causes a large and statistically significantincrease in the share of individuals who self-report foreign income, while the moralletters on average has no effect on the extensive margin.

In Figure 2, we report average self-reported foreign income for the group thatreports a positive amount in each treatment. We observe that the conditional av-erage foreign income reported is about 50% higher in the Moral treatments thanin the Base treatment, while we see no difference between the Base treatment andthe Detection treatment. In fact, the conditional average foreign income reported islower in the Base treatment and the Detection treatment than in the No letter group.When interpreting these comparisons, it is important to keep in mind how the treat-ments affected the extensive margin, as reported in Result 3. The moral treatmentshad a minor effect on the extensive margin, which means that the large increase inthe conditional average foreign income can be interpreted as the moral treatmentshaving a significant effect on the intensive margin. The moral treatments thus ap-pear primarily to have motivated taxpayers who already report some foreign incometo reduce their misreporting. For the Base and Detection treatments, however, it isharder to identify the effect on the intensive margin, since both these treatments alsohad a large effect on the extensive margin. The overall effect on the conditional av-erage foreign income is thus a result of two forces potentially working in oppositedirections: a selection effect where more individuals report a positive (and possiblysmall) amount in these treatments, and an effect on the intensive margin.

[ Figure 2 about here. ]

In tables 4 and 5, we observe that the inclusion of the background variablesdoes not change the estimated treatment effects, but some of the variables are pre-dictive for how much is self-reported in the follow-up year. We observe that thereis a highly significant positive association between self-reported foreign income atbaseline and in the follow-up year: taxpayers who self-report foreign income in thebaseline are more likely to self-report foreign income in the follow-up year and theamount self-reported is increasing in the amount they self-reported at the baseline.We also observe that the foreign income recorded in the ACA-reports is positivelyassociated with self-reported foreign income, but we only find a statistically signif-icant relationship at the extensive margin. This may reflect that the level of foreignincome in the ACA-reports is a noisy measure of actual foreign income, since thesereports only contain information from some countries and may even for these coun-tries have an imprecise measure of the taxpayer’s actual income. Interestingly, wefind that females and older people tend to report higher levels of foreign income

16

and are also more likely to report foreign income, which is consistent with the com-mon finding that these personal characteristics are positively associated with be-ing morally motivated (Andreoni, Erard, and Feinstein, 1998; Cappelen, Nygaard,Sørensen, and Tungodden, 2015). We also observe that individuals with higher in-come or greater wealth self-report higher foreign income, while we do not find thatNorwegian citizens are significantly different in their self-reporting behavior thannon-Norwegian citizens.

6.2 Heterogeneity analysisIn Table 6, we report estimated treatment effects by subgroup based on equation (4),where we focus on the Moral treatments combined (panel A) and the Detectiontreatment (panel B) compared to the Base treatment.14 The most striking featureof this analysis is the consistency in the estimated treatment effects: for all sub-groups, we observe that the moral letters and the detection letter increase the levelof self-reported foreign income. Not surprisingly, since we here look at smallersubsamples, the estimated effect is not statistically significant for all subgroups, butthe robust pattern speaks of these letters having increased self-reported foreign in-come. Similarly, also for the extensive margin, we find the same patterns acrosssubgroups as in the main analysis: the moral letters have typically a very small orno effect on the extensive margin, while the detection letter has a large and highlystatistically significant effect in all subgroups. Overall, the heterogeneity analysisclearly demonstrates that our main findings are robust across subgroups.

[ Table 6 about here. ]

The fact that the patterns are very similar across subgroups are also reflected inmost interactions between treatments and the background characteristics not beingsignificant, with the exception of the interaction between the Moral treatment andhigh socioeconomic status for amount reported (p = 0.016, panel A) and betweenthe Detection treatment and being a Norwegian citizen (p = 0.010, panel B) orabove 60 years (p < 0.001, panel B).

In the columns “Positive base”, we report the interaction between the treatmentand an indicator variable for whether an individual self-reported a positive foreignincome in the baseline year. In line with the finding that the Moral treatmentsprimarily worked on the intensive margin, we observe in panel A that the effect

14In Table A5 in Appendix A, we report the heterogeneity analysis for each of the two Moraltreatments.

17

of moral suasion on amount reported is particularly strong for the group that self-reported a positive amount in the baseline year (p = 0.026). In contrast, the Moraltreatment had only a marginal positive effect on the taxpayers that did not report anyforeign income in the baseline year, both in terms of the amount reported and sharereporting a positive amount. The pattern is strikingly different for the Detectiontreatment. In panel B, we observe that the detection letter worked on the extensivemargin for both groups, but particularly for those who did not report any foreignincome in the baseline year (p < 0.001). The Detection treatment also caused anincrease in the amount reported in both groups, but this effect is only statisticallysignificant for the taxpayers that did not report any foreign income in the baselineyear (p = 0.003).

Result 4: The effects of the moral letters and the detection letter are robustacross subgroups, with few significant interaction effects between subgroups andtreatment. The heterogeneity analysis provides evidence consistent with the moralletters strongly affecting the intensive margin and the detection letter strongly af-fecting the extensive margin of tax payer behavior.

6.3 Long-term evidenceIn this part, we study the self-reporting behavior of the taxpayers in our sample in2014, when they had to self-report their foreign income in the pre-populated taxreturn for 2013. The deadline was again in week 18, which means their choice ofhow much to self-report for 2013 happened more than one year after they receivedthe treatment letters.

In the long term, the treatment letters may not only affect the choice of howmuch to self-report, but also the choice of how much income-generating activity tohave abroad. We have shown that the treatment letters in the follow-up year causedmore taxpayers to self-report foreign income, which again may imply that they findit less attractive to earn money abroad (because they expect to pay more taxes onforeign income).15

In Figure 3, we provide an overview of our long-term findings.16 In panel A,

15However, it should be kept in mind the taxpayers received the letter in week 15 of 2013, whichmeans that they only to a limited extent had the opportunity to change the extent to which theyearned income abroad in 2013 as a response to the intervention.

16Tables A6, A7, and A8 in Appendix A report regression estimates for the long-term analysis.In Table A8, we show that our findings are robust to the removal of the 700 individuals in the Noletter group that participated in the independent validation survey of the letters that took place early2014, and to the removal of the individuals who were most likely to be followed up by the taxauthorities in 2013. The latter analysis is based on communication with the NTA, who provided us

18

we observe that the average self-reported foreign income in the Moral treatments issomewhat higher than in the other treatments, but this difference is not statisticallysignificant. In panel B, however, we observe that even in the long term, there isa large and statistically significant effect of the detection letter on the extensivemargin: the share of taxpayers reporting a positive foreign income increases from25% in the Base treatment to 32.6% in the Detection treatment (p < 0.001, column(6) in Table A7). We also observe that the base letter itself has some effect on theextensive margin in the long-run compared to not receiving a letter (p = 0.070), butwe do not find any effect of the moral letters (p = 0.684).

[ Figure 3 about here. ]

Interestingly, in panel C, we observe that the average self-reported foreign in-come for the group that reports a positive amount is significantly lower in the De-tection treatment than in the other treatments. This is consistent with the detectionletter making it less attractive to earn income abroad. In line with what we shouldexpect from panels A and B, we do not observe any other significant differences inpanel C.

Overall, the long-term data provide evidence of the Detection treatment havinga lasting effect on taxpayer behavior, by significantly increasing the share of tax-payers who self-report a positive foreign income. In contrast, the Moral treatmentsprimarily seem to have an effect in the short term.

Result 5: The detection letter has a significant long-term effect on the extensivemargin, while we do not find any significant long-term effects of the moral letters.

The difference in long-term effects of the detection letter and the moral lettersmay speak to the underlying mechanisms driving the initial effects observed in thefollow-up year. The Moral treatments may primarily have made moral argumentssalient when the taxpayers where due to report in the follow-up year, without caus-ing a fundamental change in the preferences of the individual and therefore notchanging their long-term behavior. The Detection treatment, on the other hand,may have caused the taxpayers to update their beliefs about the detection probabil-ity, and these updated beliefs may have been sustained also in the long term.

with information about their auditing rules. Note that according to our agreement with the NTA, notax payer was followed up before they had submitted their tax return in the follow-up year. Hence,auditing from the NTA could only potentially affect our long-term findings.

19

7 ConclusionsOur study shows that tax administrations should consider a rich set of instrumentsin the fight against tax evasion. A simple and cheap field intervention using let-ters increased the amount of self-reported foreign income by around 140 millionNOK (approximately 25 million USD) in the follow-up year. The intervention alsocleanly identified that both moral motivation and economic incentives play a cru-cial role in shaping taxpayer behavior. In line with the increasing focus among taxadministrators on building a tax morale in society (Luttmer and Singhal, 2014), wefind a large effect of moral suasion, of the same size as the effect of including asentence that increases the perceived probability of detection. However, moral ap-peals and detection probability influence tax behavior in different ways. The moralappeals mainly work on the intensive margin, while increasing the detection prob-ability mainly works on the extensive margin. We also report long-term effects ofthe intervention, where we show that the detection letter has a large effect on theextensive margin even one year after the taxpayer received the letter, while thereare no statistically significant long-term effects of the moral letters.

The long-term findings show that it is important to distinguish between (at least)two different mechanisms when considering how moral suasion may reduce tax eva-sion. In our study, it appears that moral suasion mainly worked by making the moralargument salient when the taxpayer made the decision of how much to report, butdid not work at a more fundamental level by increasing the weight taxpayers assignto the moral cost of misreporting (since there was no effect of the moral letters inthe long run). An important question for future research is how tax administrationsmay contribute to a more fundamental change in the tax morale in society, whichmost likely requires more extensive interventions than the use of letters. One possi-bility explored by the Norwegian tax authorities is the launch of training programstargeting adolescents on why society needs taxes. It is by now well established thatadolescence is a period of moral development (Almas, Cappelen, Sørensen, andTungodden, 2010; Fehr et al., 2013), and such interventions thus have the potentialto shape the tax morale of future taxpayers.

Our study also demonstrates that the detection probability plays a critical rolefor tax compliance, and another important avenue for future research would be tostudy how the moral motive and the detection motive interact in shaping moralbehavior. Is there crowding out of moral motivation among taxpayers when taxadministrations primarily focus on detection probability and penalty rates (Gneezyand Rustichini, 2000)? Moreover, the fact that the detection letter and the moralletters worked at different margins, shows that the context is important when con-sidering different strategies for increasing tax compliance. A focus on tax moralewill only work when tax payers consider it morally wrong to cheat on taxes, but may

20

then have significant impact by making tax payers report income that it is impos-sible to detect for the tax administration. A focus on detection probability is likelyto increase tax compliance also in settings where tax subjects are not morally moti-vated, but may cause a more narrow response where tax payers only start reportingincome that they find likely to be detected by the tax administration.

Finally, the study contributes to the broader discussion in economics about theimportance of moral motivation, by showing that moral motives not only matters inthe lab, but also in field settings involving large stakes. A simple moral messagecaused the taxpayers to self-report a significantly larger amount of foreign income,which illustrates the power of moral motivation in shaping human behavior.

ReferencesAllingham, Michael G. and Agnar Sandmo (1972). “Income tax evasion: A theo-

retical analysis,” Journal of Public Economics, 1(3-4): 323–338.

Alm, James, Gary H. McClelland, and William D. Schulze (1992). “Why do peoplepay taxes?” Journal of Public Economics, 48(1): 21–38.

Almas, Ingvild, Alexander W. Cappelen, Erik Ø. Sørensen, and Bertil Tungod-den (2010). “Fairness and the development of inequality acceptance,” Science,328(5982): 1176–1178.

Andreoni, James, Brian Erard, and Jonathan Feinstein (1998). “Tax compliance,”Journal of Economic Literature, 36(2): 818–860.

Andreoni, James and John Miller (2002). “Giving according to GARP: An experi-mental test of the consistency of preferences for altruism,” Econometrica, 70(2):737–753.

Ariel, Barak (2012). “Deterrence and moral persuasion effects on corporate taxcompliance: Findings from a randomized controlled trial,” Criminology, 50(1):27–69.

Artavanis, Nikolaos, Adair Morse, and Margarita Tsoutsoura (2015). “Measuringincome tax evasion using bank credit: Evidence from Greece,” mimeo, Univer-sity of California, Berkeley.

Blumenthal, Masha, Charles Christian, and Joel Slemrod (2001). “Do normativeappeals affect tax compliance? Evidence from a controlled experiment in Min-nesota,” National Tax Journal, 54(1): 125–138.

21

Bo, Ernesto Dal and Pedro Dal Bo (2009). ““Do the right thing:” The effects ofmoral suasion on cooperation,” Journal of Public Economics, 117(1): 28–38.

Bolton, Gary E. and Axel Ockenfels (2000). “ERC: A theory of equity, reciprocity,and competition,” American Economic Review, 90(1): 166–193.

Bryan, Christopher J., Gabrielle S. Adams, and Benoıt Monin (2012). “When cheat-ing would make you a cheater: Implicating the self prevents unethical behaviour,”Journal of Experimental Psychology: General, 142(4): 1001–1005.

Cappelen, Alexander W., Astri Drange Hole, Erik Ø. Sørensen, and Bertil Tun-godden (2007). “The pluralism of fairness ideals: An experimental approach,”American Economic Review, 97(3): 818–827.

Cappelen, Alexander W., Karl O. Moene, Erik Ø. Sørensen, and Bertil Tungod-den (2013). “Needs versus entitlements: An international fairness experiment,”Journal of the European Economic Association, 11(3): 574–598.

Cappelen, Alexander W., Knut Nygaard, Erik Ø. Sørensen, and Bertil Tungodden(2015). “Social preferences in the lab: A comparison of students and a represen-tative population,” Scandinavian Journal of Economics, 117(4): 1306–1326.

Carrillo, Paul E., M. Shahe Emran, and Anita Rivadeneira (2012). “Do cheatersbunch together? Profit taxes, withholding rates and tax evasion,” Working paper.

Casaburi, Lorenzo and Ugo Troiano (2016). “Ghost-house busters: The electoralresponse to a large anti-tax evasion program,” Quarterly Journal of Economics,131(1): 273–314.

Castro, Lucio and Carlos Scartascini (2015). “Tax compliance and enforcement inthe pampas – evidence from a field experiment,” Journal of Economic Behaviorand Organization, 116(1): 65–82.

Cebula, Richard J. and Edgar L. Feige (2012). “America’s unreported economy:measuring the size, growth and determinants of income tax evasion in the U.S.”Crime, Law and Social Change, 57(3): 265–285.

Charness, Gary and Matthew Rabin (2002). “Understanding social preferences withsimple tests,” Quarterly Journal of Economics, 117(3): 817–869.

Coleman, Stephen (1996). “The Minnesota income tax compliance experiment:State tax results,” Mimeo, Minnesota Department of Revenue.

22

Coleman, Stephen (2007). “The Minnesota income tax compliance experiment:Replication of the social norms experiment,” Mimeo, Metropolitan State Uni-versity.

Del Carpio, Lucia (2013). “Are the neighbours cheating? Evidence from a socialnorm experiment on property taxes in Peru,” Working paper.

Dwenger, Nadja, Henrik Kleven, Imran Rasul, and Johannes Rincke (2016). “Ex-trinsic and intrinsic motivations for tax compliance: Evidence from a field exper-iment in Germany,” American Economic Journal: Economic Policy, 8(3): 203–232.

Engelmann, Dirk and Martin Strobel (2004). “Inequality aversion, efficiency, andmaximin preferences in simple distribution experiments,” American EconomicReview, 94(4): 857–869.

Erat, Sanjiv and Uri Gneezy (2012). “White lies,” Management Science, 58(4):723–733.

Fehr, Ernst and Simon Gachter (1998). “Reciprocity and economics: the economicimplications of Homo Reciprocans,” European Economic Review, 42(3-5): 845–859.

Fehr, Ernst, Daniela Glatzle-Rutzler, and Matthias Sutter (2013). “The develop-ment of egalitarianism, altruism, spite and parochialism in childhood and adoles-cence,” European Economic Review, 64(1): 369–383.

Fehr, Ernst and Klaus M. Schmidt (1999). “A theory of fairness, competition andcooperation,” Quarterly Journal of Economics, 114(3): 817–868.

Fellner, Gerlinde, Rupert Sausgruber, and Christian Traxler (2013). “Testing en-forcement strategies in the field: threat, moral appeals and social information,”Journal of the European Economic Association, 11(3): 634–660.

Fisman, Raymond and Shang-Jin Wei (2004). “Tax rates and tax evasion: evidencefrom “missing imports” in China,” Journal of Political Economy, 112(2): 471–500.

Gneezy, Uri (2005). “Deception: The role of consequences,” American EconomicReview, 95(1): 384–394.

Gneezy, Uri and Aldo Rustichini (2000). “Pay enough or don’t pay at all,” QuarterlyJournal of Economics, 115(3): 791–810.

23

Hallsworth, Michael (2014). “The use of field experiments to increase tax compli-ance,” Oxford Review of Economic Policy, 30(4): 658–679.

Hallsworth, Michael, John A. List, Robert D. Metcalf, and Ivo Vlaev (2017). “Thebehavioralist as tax collector: Using natural field experiments to enhance taxcompliance,” Journal of Public Economics, 148(1): 14–31.

Kleven, Henrik Jacobsen (2014). “How can Scandinavians tax so much?” Journalof Economic Perspectives, 28(4): 77–98.

Kleven, Henrik Jacobsen, Martin B. Knudsen, Claus Thustrup Kreiner, Søren Ped-ersen, and Emmanuel Saez (2011). “Unwilling or unable to cheat? Evidencefrom a tax audit experiment in Denmark,” Econometrica, 79(3): 651–692.

Konow, James (2000). “Fair Shares: Accountability and Cognitive Dissonance inAllocation Decisions,” American Economic Review, 90(4): 1072–1091.

Lamberton, Cait, Jan-Emmanuel De Neve, and Michael I. Norton (2014). “Elicitingtaxpayer preferences increases tax compliance,” Working Paper.

Levitt, Steven D. and John A. List (2007). “What do laboratory experiments mea-suring social preferences reveal about the real world?” Journal of EconomicPerspectives, 21(2): 153–174.

Luttmer, E. P. and Monica Singhal (2014). “Tax morale,” Journal of Economic Per-spectives, 28(4): 149–168.

Rincke, Johannes and Christian Traxler (2010). “Enforcement spillovers,” Reviewof Economics and Statistics, 93(4): 1224–1234.

Sandmo, Agnar (2005). “The theory of tax evasion: A retrospective view,” NationalTax Journal, 58(4): 643–663.

Sandmo, Agnar (2012). “An evasive topic: Theorizing about the hidden economy,”International Tax and Public Finance, 19(1): 5–24.

Slemrod, Joel (2007). “Cheating ourselves: The economics of tax evasion,” Journalof Economic Perspectives, 21(1): 25–48.

Slemrod, Joel and Shlomo Yitzhaki (2002). “Tax avoidance, evasion, and adminis-tration,” in Alan J. Auerbach and Martin Feldstein (eds.), “Handbook of PublicEconomics,” volume 3, Elsevier, pp. 1423–1470.

24

010

000

2000

030

000

Mea

n ±

s.e

.

No let

ter

Base

Mor

al

Detec

tion

A. Level of reported foreign income

0.0

5.1

.15

.2.2

5.3

Sha

re ±

s.e

.

No let

ter

Base

Mor

al

Detec

tion

B. Share reporting foreign income

Figure 1: Self-reported foreign incomePanel A shows the average amount of self-reported foreign income (in NOK) and Panel B shows theshare of taxpayers who self-report a positive foreign income, by treatment.

25

050

000

1000

0015

0000

Mea

n ±

s.e

.

No let

ter

Base

Mor

al

Detec

tion

Figure 2: Self-reported foreign income – conditional on it being positiveThe figure shows the average amount of self-reported foreign income (in NOK) for the subset oftaxpayers that report a positive foreign income.

26

27

020

000

4000

0M

ean

± s

.e.

No let

ter

Base

Mor

al

Detec

tion

A. Level of reported foreign income

0.1

.2.3

.4S

hare

± s

.e.

No let

ter

Base

Mor

al

Detec

tion

B. Share reporting foreign income

010

0000

Mea

n ±

s.e

.

No let

ter

Base

Mor

al

Detec

tion

C. Level conditional on positive

Figure 3: Long-term (2013) self-reported foreign incomeThe figure shows self-reported foreign income for the tax year 2013, which was self-reported morethan one year after the intervention. Panel A shows the average amount of self-reported foreignincome (in NOK), Panel B shows the share of taxpayers who self-report a positive foreign income,and Panel C shows the average amount of self-reported foreign income (in NOK) for the subset oftaxpayers that report a positive foreign income.

Table 1: Descriptive statistics on the samples: General

Samples ACA-report

In ACA Generalreports population Incorrect Correct

Share Norwegian citizen 0.522 0.836 0.550 0.503Share citizen of other Nordic country 0.456 0.039 0.431 0.474Share female 0.445 0.502 0.455 0.437Mean age 53.4 49.8 58.4 49.9Share older than 60 years old 0.429 0.289 0.566 0.332Share self-employed 0.117 0.084 0.095 0.133

Observations 37 897 215 956 15 708 22 189

The general population refers to a 5% sample of the population in the Norwegian tax records that donot have an ACA-report. For the individuals in the Norwegian tax records that have an ACA-report,“Incorrect” denotes the set of individuals who have self-reported foreign income incorrectly and“Correct” denotes the set of individuals who have self-reported foreign income incorrectly.

28

29

Table 2: Income, wealth, and misreporting in baseline year

Samples ACA-report

In ACA Generalreports population Incorrect Correct

A. Taxable income:mean 360 628 272 616 299 838 403 619Q25 119 834 110 447 97 199 147 551Q50 234 809 215 354 182 845 274 685Q75 419 110 345 076 347 295 458 413

B. Taxable wealth:mean 1 330 938 462 820 1 530 805 1 189 590Q25 0 0 0 0Q50 43 248 63 35 277 35 277Q75 609 583 325 706 577 269 577 269

C. ACA-reports of foreign income:mean 44 902 56 280 36 852Q25 519 7 509 105Q50 6 560 18 987 868Q75 29 073 48 670 12 284

D. Estimate of misreporting:mean 8 866 28 533 -5 049Q25 16 4 187 3Q50 985 14 209 154Q75 13 556 36 732 948

Panel A and B refer to (taxable) income and wealth in the baseline year 2011. Panel C shows to-tal foreign income recorded in the ACA-reports, while panel D shows misreported foreign income(calculated by the difference between the foreign income in the ACA-reports and self-reported for-eign income). Qx refers to the x-percentile in the relevant group. The general population refers toa 5% sample of the population in the Norwegian tax records that do not have an ACA-report. Forthe individuals in the Norwegian tax records that have an ACA-report, “Incorrect” denotes the setof individuals who have self-reported foreign income incorrectly and “Correct” denotes the set ofindividuals who have self-reported foreign income incorrectly.

Tabl

e3:

Ove

rvie

wov

ertr

eatm

ents

Trea

tmen

tnam

eTr

eatm

entd

escr

iptio

n

No

lett

erD

idno

trec

eive

any

lette

r.

Bas

eG

ener

alin

form

atio

nle

tter.

Fair

ness

Bas

ele

tter+

the

follo

win

gse

nten

cead

ded

toth

efir

stpa

ragr

aph:

“The

grea

tmaj

ority

repo

rtin

form

atio

nab

outt

heir

inco

me

and

asse

tsin

Nor

way

corr

ectly

and

com

plet

ely.

Inor

der

totr

eata

llta

xpay

ers

fair

ly,i

tis

ther

efor

eim

port

antt

hatf

orei

gnin

com

ean

dfo

reig

nas

sets

are

repo

rted

inth

esa

me

man

ner.”

Soci

etal

Ben

efits

Bas

ele

tter+

the

follo

win

gse

nten

cead

ded

toth

efir

stpa

ragr

aph:

“You

rta

xpa

ymen

tcon

trib

utes

toth

efu

ndin

gof

publ

icly

finan

ced

serv

ices

ined

ucat

ion,

heal

than

dot

her

impo

rtan

tse

ctor

sof

soci

ety”

.Tw

oad

ditio

nalt

reat

men

tsin

clud

edan

atta

chm

entv

isua

lisin

gpu

blic

serv

ices

finan

ced

thro

ugh

taxe

s(w

ithou

tthe

base

lette

r/in

com

bina

tion

with

the

base

lette

r).

Det

ectio

nB

ase

lette

r,bu

tthe

first

sent

ence

(“T

heN

orw

egia

nec

onom

yis

beco

min

gm

ore

inte

rnat

iona

lised

and

anin

crea

sing

num

bero

fNor

weg

ian

taxp

ayer

sre

ceiv

ein

com

ean

dha

veas

sets

abro

ad”)

isre

plac

edby

the

follo

win

gse

nten

ce:

“The

tax

adm

inis

trat

ion

has

rece

ived

info

rmat

ion

that

you

inpr

evio

usye

ars

have

had

inco

me

and/

oras

sets

abro

ad.”

30

31

Table 4: Treatment effects on self-reported foreign income(1) (2) (3) (4) (5) (6)

No letter -8874.9∗∗∗ -10008.1∗∗∗ -3188.0∗ -4339.1 -3188.0∗ -4339.6(2184.5) (2767.4) (1643.1) (2734.8) (1643.1) (2735.2)

Fairness 15158.5∗ 10372.1∗∗

(8860.6) (4041.9)

Societal Benefits 5180.9∗∗ 6345.5∗∗

(2596.8) (2989.0)

Detection 9199.6∗∗ 10351.3∗∗ 9199.6∗∗ 10351.6∗∗

(4385.6) (4702.7) (4385.5) (4702.8)

Moral 7671.1∗∗ 7350.5∗∗∗

(3010.7) (2772.9)

Foreign income baseline 0.42∗∗ 0.42∗∗ 0.42∗∗

(0.20) (0.20) (0.20)

ACA-report baseline 0.00089 0.00091 0.00091(0.0035) (0.0035) (0.0035)

ACA-report follow-up 0.010 0.010 0.010(0.0073) (0.0073) (0.0073)

Female 6791.0∗ 6782.8∗ 6777.1∗

(3640.9) (3634.1) (3633.9)

Age > 60 yrs 9167.8∗ 9148.3∗ 9144.6∗

(5207.1) (5209.7) (5207.6)

Norwegian citizen 2702.2 2733.2 2741.5(3294.8) (3307.6) (3306.1)

High SES 4591.3 4615.8 4595.0(5275.6) (5262.7) (5274.6)

Constant 17029.8∗∗∗ -6706.3 11342.9∗∗∗ -12387.4 11342.9∗∗∗ -12380.8(1752.6) (6673.8) (999.6) (7537.0) (999.6) (7535.5)

F-test p on Moral treatments being equal: 0.27 0.32Observations 15708 15708 15708 15708 15708 15708R2 0.000 0.230 0.001 0.231 0.001 0.231

The table reports regressions based on equation 3, where the dependent variable is the amount for-eign income self-reported for the follow-up year. In columns (1) - (2), the estimated effects arerelative to the the pooled sample of all treatment groups; in columns (3)-(6), the estimated effectsare relative to the Base treatment. The indicator variables “No letter”, “Fairness”, “Societal Bene-fits”, and “Detection” take the value one if the taxpayer is in the respective treatment. The indicatorvariable “Moral” is one if the taxpayer is in the Fairness treatment or the Social Benefits treat-ment. The reported F-test p is for the hypothesis that all Moral treatments have the same effect.Columns (2), (4), and (6) include the following controls: the amount of self-reported foreign in-come for the baseline year, the amount of foreign income recorded in the ACA-reports for 2011and 2012, gender, age (an indicator variable taking the value one if the taxpayer is more than 60years), an indicator variable taking the value one if the tax payer is a Norwegian citizen, and anindicator variable of socio-economic status taking the value one if the taxpayer is in the upper 25%of the income and wealth distribution in the baseline year. Robust standard errors in parentheses.(∗ : p < 0.1,∗∗ : p < 0.05,∗∗∗ : p < 0.01).

32

Table 5: Treatment effects on self-reporting of any foreign income(1) (2) (3) (4) (5) (6)

No letter -0.11∗∗∗ -0.11∗∗∗ -0.086∗∗∗ -0.083∗∗∗ -0.086∗∗∗ -0.083∗∗∗

(0.0079) (0.0076) (0.0095) (0.0091) (0.0095) (0.0091)

Fairness 0.031∗∗∗ 0.030∗∗∗

(0.011) (0.011)

Societal Benefits 0.0086 0.0056(0.0082) (0.0079)

Detection 0.13∗∗∗ 0.13∗∗∗ 0.13∗∗∗ 0.13∗∗∗

(0.012) (0.012) (0.012) (0.012)

Moral 0.014∗ 0.012(0.0078) (0.0075)

Positive foreign income baseline 0.16∗∗∗ 0.16∗∗∗ 0.16∗∗∗

(0.0073) (0.0073) (0.0073)

ACA-report baseline 0.022∗ 0.022∗ 0.022∗

(0.012) (0.013) (0.013)

ACA report follow-up 0.031 0.029 0.029(0.023) (0.024) (0.023)

Female 0.052∗∗∗ 0.052∗∗∗ 0.052∗∗∗

(0.0063) (0.0063) (0.0063)

Age> 60 yrs 0.21∗∗∗ 0.21∗∗∗ 0.21∗∗∗

(0.0075) (0.0075) (0.0075)

Norwegian citizen 0.00033 0.00014 0.00019(0.0069) (0.0068) (0.0068)

High SES 0.037∗∗∗ 0.038∗∗∗ 0.038∗∗∗

(0.0070) (0.0070) (0.0070)

Constant 0.22∗∗∗ -0.0035 0.20∗∗∗ -0.029∗∗∗ 0.20∗∗∗ -0.029∗∗∗

(0.0035) (0.0065) (0.0063) (0.0081) (0.0063) (0.0081)

F-test p on Moral treatments being equal: 0.038 0.021Observations 15708 15708 15708 15708 15708 15708R2 0.008 0.091 0.019 0.102 0.019 0.102

The table reports regressions based on equation 3, where the dependent variable is an indicatorvariable taking the value one if the taxpayer self-reports any foreign income reported for the follow-up year. In columns (1) - (2), the estimated effects are relative to the the pooled sample of alltreatment groups; in columns (3)-(6), the estimated effects are relative to the Base treatment. Theindicator variables “No letter”, “Fairness”, “Societal Benefits”, and “Detection” take the value one ifthe taxpayer is in the respective treatment. The indicator variable “Moral” is one if the taxpayer is inthe Fairness treatment or the Social Benefits treatment. The reported F-test p is for the hypothesisthat all Moral treatments have the same effect. Columns (2), (4), and (6) include the followingcontrols: an indicator variable taking the value one if the taxpayer self-reported any foreign incomefor the baseline year, the amount of foreign income recorded in the ACA-reports for 2011 and 2012(scaled in units of 1 000 000 NOK), gender, age (an indicator variable taking the value one if thetaxpayer is more than 60 years), an indicator variable taking the value one if the tax payer is aNorwegian citizen, and an indicator variable of socio-economic status taking the value one if thetaxpayer is in the upper 25% of the income and wealth distribution in the baseline year. Robuststandard errors in parentheses. (∗ : p < 0.1,∗∗ : p < 0.05,∗∗∗ : p < 0.01).

Tabl

e6:

Het

erog

enei

tyin

how

Mor

alan

dD

etec

tion

trea

tmen

tsw

ork

A.H

eter

ogen

eity

-Mor

altr

eatm

ents

:L

evel

repo

rted

(in

NO

K)

Rep

ortin

gpo

sitiv

eam

ount

Citi

zen

Abo

ve60

Fem

ale

Hig

hSE

SPo

sitiv

eba

seC

itize

nA

bove

60Fe

mal

eH

igh

SES

Posi

tive

base

Mor

al93

16.5∗∗

9626

.870

73.3∗

634.

721

32.3∗∗

0.00

720.

0018

0.02

3∗∗

0.00

920.

015∗

(471

7.6)

(590

5.8)

(414

1.6)

(162

9.1)

(995

.1)

(0.0

10)

(0.0

092)

(0.0

097)

(0.0

092)

(0.0

093)

Gro

up×

Mor

al-3

518.

5-4

059.

466

8.2

1484

9.2∗∗

1438

2.9∗

0.00

870.

018

-0.0

240.

0063

-0.0

084

(601

8.2)

(616

4.7)

(571

7.9)

(613

5.6)

(745

8.9)

(0.0

15)

(0.0

15)

(0.0

15)

(0.0

15)

(0.0

15)

Gro

up70

92.0

1273

4.4∗∗

7892

.6∗

-609

2.3

1241

2.7∗∗∗

-0.0

065

0.20∗∗∗

0.07

4∗∗∗

0.03

5∗∗∗

0.16∗∗∗

(474

5.9)

(588

4.0)

(471

0.0)

(586

7.8)

(350

5.7)

(0.0

12)

(0.0

13)

(0.0

12)

(0.0

13)

(0.0

13)

Trea

tmen

teff

ecto

n‘g

roup

’57

98.1

5567

.3∗∗∗

7741

.6∗∗

1548

3.9∗∗∗

1651

5.1∗∗

0.01

60.

020∗

-0.0

013

0.01

60.

0070

(359

9.7)

(190

6.5)

(390

0.3)

(594

4.7)

(740

4.4)

(0.0

11)

(0.0

11)

(0.0

12)

(0.0

12)

(0.0

12)

Obs

erva

tions

1176

811

768

1176

811

768

1176

811

768

1176

811

768

1176

811

768

B.H

eter

ogen

eity

-Det

ectio

ntr

eatm

ent:

Lev

elre

port

ed(i

nN

OK

)R

epor

ting

posi

tive

amou

nt

Citi

zen

Abo

ve60

Fem

ale

Hig

hSE

SPo

sitiv

eba

seC

itize

nA

bove

60Fe

mal

eH

igh

SES

Posi

tive

base

Det

ectio

n14

298.

611

417.

855

20.4

2475

.3∗

6114

.7∗∗∗

0.10

1∗∗∗

0.06

56∗∗∗

0.13

5∗∗∗

0.11

9∗∗∗

0.17

2∗∗∗

(893

2.3)

(902

7.6)

(346

4.2)

(144

9.6)

(204

4.9)

(0.0

162)

(0.0

146)

(0.0

154)

(0.0

148)

(0.0

152)

Gro

up×

Det

ectio

n-8

609.

9-3

405.

490

36.8

1565

4.1

7707

.60.

0599∗∗

0.12

6∗∗∗

-0.0

0326

0.03

41-0

.091

8∗∗∗

(943

3.3)

(955

2.1)

(926

8.8)

(966

0.7)

(103

10.8

)(0

.023

3)(0

.022

9)(0

.023

7)(0

.023

9)(0

.023

9)

Gro

up83

5.5

5082

.4∗∗

-184

0.6

4499

.8∗

1576

5.3∗∗∗

-0.0

147

0.19

5∗∗∗

0.07

18∗∗∗

0.03

36∗∗

0.17

2∗∗∗

(238

7.8)

(227

1.3)

(176

4.9)

(233

0.6)

(287

2.1)

(0.0

131)

(0.0

138)

(0.0

124)

(0.0

132)

(0.0

140)

Trea

tmen

teff

ecto

n‘g

roup

’56

88.7∗

8012

.4∗∗∗

1455

7.2∗

1812

9.4∗

1382

2.3

0.16

1∗∗∗

0.19

1∗∗∗

0.13

2∗∗∗

0.15

3∗∗∗

0.08

00∗∗∗

(309

0.8)

(292

2.7)

(868

4.7)

(952

7.0)

(101

38.6

)(0

.016

7)(0

.017

7)(0

.018

0)(0

.018

8)(0

.018

4)

Obs

erva

tions

5919

5919

5919

5919

5919

5919

5919

5919

5919

5919

The

tabl

ere

port

sre

gres

sion

estim

ates

base

don

the

equa

tion

4,w

ithth

esa

me

seto

fco

ntro

lsas

inTa

ble

4an

dTa

ble

5.C

olum

nhe

ader

ssh

owth

ein

dica

tor

vari

able

that

isus

edto

defin

eth

ein

dica

tor

vari

able

“Gro

up”,

whe

re“G

roup

”ta

kes

the

valu

eon

eif

the

indi

cato

rva

riab

lein

the

head

ing

ofth

ere

spec

tive

colu

mn

take

sth

eva

lue

one:

“Citi

zen”

(the

taxp

ayer

isa

Nor

weg

ian

citiz

en),

“Hig

hSE

S”(t

heta

xpay

eris

abov

e60

year

s),“

Fem

ale”

(the

taxp

ayer

isa

fem

ale)

,“H

igh

SES”

(the

taxp

ayer

isin

the

uppe

r25

%of

the

inco

me

and

wea

lthdi

stri

butio

nin

the

base

line

year

),an

d“P

ositi

veba

se”

(the

taxp

ayer

self

-rep

orte

dan

yfo

reig

nin

com

efo

rth

eba

selin

eye

ar).

Eff

ects

are

estim

ated

rela

tive

toth

eB

ase

trea

tmen

t.Pa

nelA

repo

rts

estim

ates

for

the

taxp

ayer

sw

how

ere

inth

eM

oral

trea

tmen

tsor

the

Bas

etr

eatm

ent,

whe

re“M

oral

”is

anin

dica

tor

vari

able

that

take

sth

eva

lue

one

ifth

eta

xpay

eris

inon

eof

the

Mor

altr

eatm

ents

.Pa

nelB

repo

rts

estim

ates

for

the

taxp

ayer

sw

how

ere

inth

eD

etec

tion

trea

tmen

tor

the

Bas

etr

eatm

ent,

whe

re“D

etec

tion”

isan

indi

cato

rva

riab

leth

atta

kes

the

valu

eon

eif

the

taxp

ayer

isin

the

Det

ectio

ntr

eatm

ent.

The

depe

nden

tva

riab

leis

the

amou

ntof

fore

ign

inco

me

self

-rep

orte

dfo

rth

efo

llow

-up

year

(lef

tpa

rt)

and

anin

dica

tor

vari

able

taki

ngth

eva

lue

one

ifth

eta

xpay

erse

lf-r

epor

ted

any

fore

ign

inco

me

for

the

follo

w-u

pye

ar(r

ight

part

).R

obus

tsta

ndar

der

rors

inpa

rent

hese

s(∗

:p<

0.1,∗∗

:p<

0.05,∗∗∗

:p<

0.01

).

33

A Appendix for online publication only: Supplemen-tary material

Figure A1: The letter sent in the Base treatment

34

35

Din skatt finansierer viktige samfunnstjenester.

Figure A2: Attachment to Moral (Societal Benefits) treatments.The attachment included in the versions of the Societal Benefits treatments that included a visualelement. The subtitle to the picture states that “Your taxes finance important public services.”

010

0020

00M

ean

± s

.e.

No let

ter

Base

Mor

al

Detec

tion

A. Deduction requested (amount)

0.0

5.1

.15

.2.2

5S

hare

± s

.e.

No let

ter

Base

Mor

al

Detec

tion

B. Any deduction requested

0.0

5.1

.15

.2.2

5S

hare

± s

.e.

No let

ter

Base

Mor

al

Detec

tion

C. Reporting foreign income,not requesting deduction

Figure A3: Applications for deductions of taxes paid abroadThe graph shows, for the follow-up year, the average amount of requested deductions in Norwegiantaxes (panel A), the share that requested deductions (panel B), and the share that reported positiveamounts of foreign income without requesting any deductions (panel C).

36

Tabl

eA

1:St

atis

tics

onba

selin

ech

arac

teri

stic

sA

mou

ntA

mou

ntSh

are

Shar

eH

igh

soci

o-Po

sitiv

eam

ount

Am

ount

inA

CA

inA

CA

Shar

eab

ove

60N

orw

egia

nec

onom

icTr

eatm

ent

nre

port

ed20

11re

port

ed20

11re

port

s20

11re

port

s20

12fe

mal

eye

ars

old

citiz

enst

atus

No

lette

r1

968

0.39

330

287

8399

543

303

0.46

00.

544

0.55

30.

441

(0.0

11)

(435

5)(3

339

0)(6

761)

(0.0

11)

(0.0

11)

(0.0

11)

(0.0

11)

Bas

e3

947

0.40

227

427

5521

144

103

0.45

00.

551

0.55

30.

454

(0.0

08)

(327

8)(3

935)

(563

2)(0

.006

)(0

.008

)0.

008)

(0.0

08)

Mor

al7

821

0.40

528

040

5197

138

097

0.45

90.

561

0.54

80.

454

(0.0

06)

(278

4)(2

071)

(263

7)(0

.06)

(0.0

06)

(0.0

05)

(0.0

06)

Det

ectio

n1

972

0.41

724

689

4785

558

312

0.44

60.

542

0.54

80.

452

(0.0

11)

(343

6)(2

777)

(20

441)

(0.0

11)

(0.0

11)

(0.0

11)

(0.0

11)

Tota

l15

708

0.40

427

746

5628

042

796

0.45

50.

554

0.55

00.

452

(0.0

04)

(175

6)(4

434)

(332

1)(0

.004

)(0

.004

)(0

.004

)(0

.004

)

Ave

rage

san

dnu

mbe

rof

obse

rvat

ions

from

the

base

line

char

acte

rist

ics

used

asco

ntro

lsin

the

mai

nre

gres

sion

spec

ifica

tions

.In

the

Mor

altr

eatm

ents

,195

2ob

serv

atio

nsar

ein

the

Fair

ness

trea

tmen

t,5

869

inth

eSo

ciet

alB

enefi

tstr

eatm

ent.

37

Tabl

eA

2:St

atis

tics

from

the

cont

actl

ogs

Test

vs.B

ase

trea

tmen

tTr

eatm

entg

roup

p-va

lues

Tota

lB

ase

Mor

alD

etec

tion

Mor

alD

etec

tion

Shar

eof

lette

rrec

ipie

nts

inth

elo

gs0.

050

0.03

60.

037

0.13

10.

710

<0.

001

Com

mun

icat

ion

with

lette

rre

cipi

ent:

Que

stio

nab

outw

hyhe

/she

rece

ives

lette

r?0.

282

0.26

10.

254

0.32

40.

888

0.18

2N

egat

ive

reac

tion

tole

tter

0.05

50.

049

0.04

10.

073

0.70

10.

364

Que

stio

nab

outr

ules

fort

axat

ion

0.34

70.

348

0.35

10.

343

0.95

10.

936

Arg

uing

abou

tleg

itim

acy

ofru

les

0.06

30.

070

0.05

80.

065

0.62

10.

832

Que

stio

nab

outh

owto

repo

rtfo

reig

nin

com

e0.

721

0.74

10.

758

0.66

80.

709

0.12

6Q

uest

ion

abou

thow

fore

ign

inco

me

isau

dite

d0.

019

0.00

70.

014

0.03

10.

540

0.12

7Q

uest

ion

abou

trep

ortin

gin

com

efr

ombe

fore

2012

0.06

20.

035

0.04

10.

099

0.76

20.

021

Cha

ract

eris

tics

ofca

ller:

Mea

nag

e67

.370

.468

.364

.40.

105

<0.

001

Shar

efe

mal

e0.

544

0.64

30.

558

0.47

50.

089

<0.

001

Shar

eN

orw

egia

nci

tizen

0.61

10.

664

0.60

50.

589

0.22

30.

138

Obs

erva

tions

700

143

294

263

The

tabl

esh

ows

shar

esof

taxp

ayer

s,ba

sed

on70

0co

ntac

tsw

ithth

ein

form

atio

nce

nter

inw

hich

calle

rsid

entifi

edw

hich

lette

rthe

yha

dre

ceiv

ed.

“Tes

tvs.

Bas

etr

eatm

ent”

repo

rts

test

sof

equa

lity

betw

een

the

resp

ectiv

etr

eatm

enta

ndth

eB

ase

trea

tmen

t.T

hep-

valu

esar

eba

sed

onPe

arso

nχ

test

son

bina

ryou

tcom

es(a

ndt-

test

sfo

rtes

tsof

mea

nag

e).

38

39

Table A3: Treatment effects: All subtreatmentsLevel reported (in NOK) Reporting positive amount

(1) (2) (3) (4)

No letter -3188.0∗ -4339.0 -0.086∗∗∗ -0.083∗∗∗

(1643.3) (2734.9) (0.0095) (0.0091)

Fairness 15158.5∗ 10372.3∗∗ 0.031∗∗∗ 0.030∗∗∗

(8861.2) (4042.0) (0.011) (0.011)

Societal Benefits (SB) 9389.6 9389.3 0.0055 -0.00073(6791.9) (6980.4) (0.011) (0.011)

Visual SB 3365.1 6107.5∗∗ 0.0066 0.0052(2492.2) (3107.5) (0.011) (0.011)

Visual and verbal SB 2931.0 3650.3 0.014 0.012(2048.5) (2688.6) (0.011) (0.011)

Detection 9199.6∗∗ 10350.8∗∗ 0.13∗∗∗ 0.13∗∗∗

(4385.9) (4703.0) (0.012) (0.012)

Baseline value 0.42∗∗ 0.16∗∗∗

(0.20) (0.0073)

ACA-report baseline 0.00091 0.022∗

(0.0035) (0.013)

ACA-report follow-up 0.010 0.029(0.0073) (0.024)

Female 6808.3∗ 0.052∗∗∗

(3649.0) (0.0063)

Age > 60 yrs 9123.2∗ 0.22∗∗∗

(5247.0) (0.0075)

Norwegian citizen 2686.8 0.00026(3281.0) (0.0068)

High SES 4637.9 0.038∗∗∗

(5278.2) (0.0070)

Constant 11342.9∗∗∗ -12368.9 0.20∗∗∗ -0.030∗∗∗

(999.7) (7535.1) (0.0063) (0.0081)

F-test p on SB 0.65 0.59 0.79 0.59Observations 15708 15708 15708 15708R2 0.001 0.231 0.019 0.102

The table reports regressions using the same specifications as in the main Tables 4 and 5, but withthe set of treatments expanded to include all sub-treatments. “Baseline value” is the baseline valueof the dependent variable. The F-test refers to the hypothesis that all coefficients for the SocietalBenefits treatments are the same. For columns (3) and (4), ACA-reports for the baseline year andfollow-up year have been scaled in units of 1 000 000 NOK. Robust standard errors in parentheses(∗ : p < 0.1,∗∗ : p < 0.05,∗∗∗ : p < 0.01).

40

Table A4: Independent survey of letters conducted in 2014Most Taxes

people finance Probabilityreport important of no Treatment Not-treatment

correctly services detection question questions

Fairness 0.053 -0.013 0.0081(0.089) (0.085) (0.091)

Societal Benefits -0.015 0.096 -0.050(0.090) (0.086) (0.090)

Detection -0.070 0.063 0.12(0.088) (0.085) (0.090)

Treated 0.089∗ -0.013(0.053) (0.048)

Female 0.0093 -0.034 0.20∗∗∗ 0.064 0.020(0.063) (0.060) (0.064) (0.052) (0.045)

Norwegian citizen 0.11 -0.034 -0.15∗∗ -0.027 -0.025(0.067) (0.064) (0.068) (0.053) (0.048)

Age -0.00013 0.0042∗∗ 0.013∗∗∗ 0.0067∗∗∗ 0.0055∗∗∗

(0.0019) (0.0018) (0.0019) (0.0015) (0.0014)Constant -0.056 -0.24∗ -0.82∗∗∗ -0.44∗∗∗ -0.34∗∗∗

(0.14) (0.13) (0.14) (0.12) (0.11)

Observations 1046 1061 948 1486 2274The table shows responses to questions in a post-intervention survey with 4000 tax subjects (700from the control group and 3300 from the group with ACA-reports that self-reportered correctly;these groups were not tracked in the survey). They were randomly assigned to receive one of fourletters (base, fairness, societal benefits, detection). After reading the letter they were asked to com-plete a survey. In addition to a set of general questions, they were asked to state the extent to whichthey agreed with the following statements (scale 1-5, disagree–agree) : A) “Most people report in-come correctly and comprehensively in their tax returns” and B) “Tax payments finance importantpublic services”. Further, they were asked about their subjective detection probability, C) “Howprobable do you believe it to be that it would be detected if you did not report foreign income (re-port a number between 0 and 100%).” 27.2% of the individuals responded to the survey, with nodifferences in response rate across treatments (Pearson’s chi-squared test, p = 0.401). In the leftpart of the table, we report responses for each of the three questions (in standard deviations), where“Fairness”, “Societal Benefits”, and “Detection” are indicator variables taking the value one if thesurvey participant received the letter used in the corresponding treatment. The effects are estimatedrelative the the response of the survey participants who received the base letter. “Female” is anindicator variable taking the value one if the survey participant is a female, “Norwegian citizen” isan indicator variable taking the value one if the survey participant is a Norwegian citizen, “Age” isthe age of the survey participant (in years). We observe that the individuals systematically respondin line with our assumptions on the issue that was highlighted in the letter they received. We alsoobserve that there is no effect of the moral letters on the subjective detection probability. The effectsare not large, but it should be kept in mind that the survey manipulation is weak and the samplemostly consists of individuals who self-reported correctly at baseline. In the right part of the panel,we show that we indeed do find a statistically significant effect of the letters working as intended.In the column “Treatment question”, we compare the response to the question that the letter theyhad received highlighted (for example the response to question C for the individuals receiving letterC) to the response to the same question by the individuals who had received the base letter, where“Treated” is an indicator variable for whether the respondent received one of the treatment lettersor the base letter. Correspondingly, in the column “Not-treatment questions”, we compare the re-sponses to the questions that the letter they had received did not highlight (for example the responseson questions A and B for the individuals receiving letter C) to the responses to the same questionsby the individuals who had received the base letter. Robust standard errors clustered on individuals(∗ : p < 0.1,∗∗ : p < 0.05,∗∗∗ : p < 0.01).

Tabl

eA

5:H

eter

ogen

eity

inho

wFa

irne

ssan

dSo

ciet

alB

enefi

tstr

eatm

ents

wor

kA

.Het

erog

enei

ty-F

airn

esst

reat

men

t:L

evel

repo

rted

(in

NO

K)

Rep

ortin

gpo

sitiv

eam

ount

Citi

zen

Abo

ve60

Fem

ale

Hig

hSE

SPo

sitiv

eba

seC

itize

nA

bove

60Fe

mal

eH

igh

SES

Posi

tive

base

Fair

ness

1626

7.6∗∗

1149

6.7

8067

.235

80.6∗

2658

.3∗∗

0.03

7∗∗

0.03

2∗∗

0.03

9∗∗∗

0.02

10.

025∗

(811

4.2)

(810

6.5)

(666

2.7)

(213

4.0)

(124

2.3)

(0.0

15)

(0.0

14)

(0.0

14)

(0.0

13)

(0.0

14)

Gro

up×

Fair

ness

-143

65.5

-564

6.1

648.

110

677.

431

625.

2-0

.013

-0.0

034

-0.0

190.

021

0.01

2(8

877.

0)(8

906.

0)(8

281.

0)(8

470.

9)(2

2203

.2)

(0.0

22)

(0.0

21)

(0.0

22)

(0.0

22)

(0.0

23)

Gro

up86

54.8∗

1657

4.7∗∗∗

8439

.8∗

-866

1.5

1201

6.8∗∗

-0.0

018

0.21∗∗∗

0.07

4∗∗∗

0.03

8∗∗∗

0.16∗∗∗

(520

4.0)

(607

3.9)

(490

5.7)

(643

0.2)

(520

8.8)

(0.0

13)

(0.0

14)

(0.0

12)

(0.0

13)

(0.0

14)

Trea

tmen

teff

ecto

n‘g

roup

’:19

02.1

5850

.6∗

8715

.4∗

1425

8.1∗

3428

3.5

0.02

40.

029∗

0.02

00.

042∗∗

0.03

7∗∗

(305

5.3)

(317

8.3)

(420

3.5)

(824

9.6)

(221

05.2

)(0

.015

)(0

.016

)(0

.017

)(0

.018

)(0

.018

)

Obs

erva

tions

5899

5899

5899

5899

5899

5899

5899

5899

5899

5899

B.H

eter

ogen

eity

-Soc

ieta

lBen

efits

trea

tmen

ts:

Lev

elre

port

ed(i

nN

OK

)R

epor

ting

posi

tive

amou

nt

Citi

zen

Abo

ve60

Fem

ale

Hig

hSE

SPo

sitiv

eba

seC

itize

nA

bove

60Fe

mal

eH

igh

SES

Posi

tive

base

Soci

etal

Ben

efits

3388

.069

81.8

4425

.7∗

-579

.119

29.8∗

-0.0

025

-0.0

083

0.01

8∗0.

0050

0.01

2(2

251.

1)(5

459.

1)(2

445.

7)(1

049.

0)(1

147.

9)(0

.011

)(0

.009

6)(0

.010

)(0

.009

7)(0

.009

8)

Gro

up×

Soci

etal

Ben

efits

4067

.0-2

435.

726

47.3

1363

9.7∗∗

8466

.10.

016

0.02

6∗-0

.026∗

0.00

23-0

.015

(469

7.0)

(567

9.4)

(538

9.4)

(569

0.7)

(622

6.3)

(0.0

16)

(0.0

15)

(0.0

16)

(0.0

16)

(0.0

16)

Gro

up18

63.1

3538

.132

6.7

3017

.614

769.

6∗∗∗

-0.0

065

0.20∗∗∗

0.07

4∗∗∗

0.03

4∗∗∗

0.16∗∗∗

(266

8.4)

(290

5.3)

(197

0.2)

(217

9.6)

(273

1.8)

(0.0

13)

(0.0

13)

(0.0

12)

(0.0

13)

(0.0

13)

Trea

tmen

teff

ecto

n‘g

roup

’:74

55.0∗

4546

.1∗∗∗

7072

.913

060.

6∗∗

1039

5.9∗

0.01

30.

017

-0.0

082

0.00

73-0

.003

1(4

281.

2)(1

715.

5)(4

887.

2)(5

613.

5)(6

183.

3)(0

.011

)(0

.012

)(0

.012

)(0

.013

)(0

.013

)

Obs

erva

tions

9816

9816

9816

9816

9816

9816

9816

9816

9816

9816

The

tabl

ere

port

sre

gres

sion

estim

ates

base

don

the

equa

tion

4,w

ithth

esa

me

seto

fcon

trol

sas

inTa

ble

4an

dTa

ble

5.C

olum

nhe

ader

ssh

owth

ein

dica

torv

aria

ble

that

isus

edto

defin

eth

ein

dica

torv

aria

ble

“Gro

up”,

whe

re“G

roup

”ta

kes

the

valu

eon

eif

the

indi

cato

rvar

iabl

ein

the

head

ing

ofth

ere

spec

tive

colu

mn

take

sth

eva

lue

one:

“Citi

zen”

(the

taxp

ayer

isa

Nor

weg

ian

citiz

en),

“Hig

hSE

S”(t

heta

xpay

eris

abov

e60

year

s),

“Fem

ale”

(the

taxp

ayer

isa

fem

ale)

,“H

igh

SES”

(the

taxp

ayer

isin

the

uppe

r25

%of

the

inco

me

and

wea

lthdi

stri

butio

nin

the

base

line

year

),an

d“P

ositi

veba

se”

(the

taxp

ayer

self

-rep

orte

dan

yfo

reig

nin

com

efo

rth

eba

selin

eye

ar).

Eff

ects

are

estim

ated

rela

tive

toth

eB

ase

trea

tmen

t.Pa

nelA

repo

rts

estim

ates

fort

heta

xpay

ers

who

wer

ein

the

Fair

ness

trea

tmen

tsor

the

Bas

etr

eatm

ent,

whe

re“F

airn

ess”

isan

indi

cato

rvar

iabl

eth

atta

kes

the

valu

eon

eif

the

taxp

ayer

isin

the

Fair

ness

trea

tmen

t.Pa

nelB

repo

rts

estim

ates

fort

heta

xpay

ers

who

wer

ein

the

Soci

etal

Ben

efits

trea

tmen

tort

heB

ase

trea

tmen

t,w

here

“Soc

ieta

lBen

efit”

isan

indi

cato

rvar

iabl

eth

atta

kes

the

valu

eon

eif

the

taxp

ayer

isin

the

Soci

etal

Ben

efits

trea

tmen

t.T

hede

pend

ent

vari

able

isth

eam

ount

offo

reig

nin

com

ese

lf-r

epor

ted

for

the

follo

w-u

pye

ar(l

eft

part

)an

dan

indi

cato

rva

riab

leta

king

the

valu

eon

eif

the

taxp

ayer

self

-rep

orte

dan

yfo

reig

nin

com

efo

rth

efo

llow

-up

year

(rig

htpa

rt).

Rob

usts

tand

ard

erro

rsin

pare

nthe

ses

(∗:p

<0.

1,∗∗

:p<

0.05,∗∗∗

:p<

0.01

).

41

42

Table A6: Long-term treatment effects on self-reported foreign income(1) (2) (3) (4) (5) (6)

No letter -3835.3 -3934.9 -1860.3 -1982.4 -1860.3 -1981.7(4319.8) (4295.7) (3601.2) (3589.0) (3601.1) (3588.8)

Fairness 912.5 -155.7(4015.1) (3621.8)

Societal Benefits 4702.7 4917.6(7683.8) (7685.1)

Detection -1138.6 -871.7 -1138.6 -872.1(3122.8) (3023.9) (3122.7) (3023.9)

Moral 3756.7 3651.4(6007.5) (6015.4)

Foreign income baseline 0.11∗∗ 0.11∗∗ 0.11∗∗

(0.046) (0.046) (0.046)

ACA-report baseline 0.0075 0.0075 0.0075(0.0099) (0.0099) (0.0099)

ACA-report follow-up 0.016 0.016 0.016(0.010) (0.010) (0.010)

Female -7501.1 -7541.2 -7534.1(4684.6) (4735.6) (4726.0)

Age > 60 yrs 20551.9∗∗∗ 20487.8∗∗∗ 20492.5∗∗∗

(3882.4) (3817.4) (3822.0)

Norwegian citizen -6216.6∗ -6178.3∗ -6188.8∗

(3633.2) (3673.7) (3661.9)

High SES 12147.0∗∗ 12125.2∗∗ 12151.4∗∗

(4997.8) (4969.4) (5003.4)

Constant 25949.9∗∗∗ 11877.7∗∗∗ 23974.9∗∗∗ 9962.5∗∗ 23974.9∗∗∗ 9954.2∗∗

(3256.6) (2458.8) (2216.3) (4458.1) (2216.3) (4465.7)

F-test p on Moral treatments being equal: 0.64 0.52Observations 15708 15708 15708 15708 15708 15708R2 0.000 0.006 0.000 0.006 0.000 0.006

The table reports regressions based on equation 3, where the dependent variable is the amount for-eign income self-reported for 2013. In columns (1) - (2), the estimated effects are relative to the thepooled sample of all treatment groups; in columns (3)-(6), the estimated effects are relative to theBase treatment. The indicator variables “No letter”, “Fairness”, “Societal Benefits”, and “Detection”take the value one if the taxpayer is in the respective treatment. The indicator variable “Moral” isone if the taxpayer is in the Fairness treatment or the Social Benefits treatment. The reported F-testp is for the hypothesis that all Moral treatments have the same effect. Columns (2), (4), and (6)include the following controls: the amount of self-reported foreign income for the baseline year, theamount of foreign income recorded in the ACA-reports for 2011 and 2012, gender, age (an indicatorvariable taking the value one if the taxpayer is more than 60 years), an indicator variable taking thevalue one if the tax payer is a Norwegian citizen, and an indicator variable of socio-economic statustaking the value one if the taxpayer is in the upper 25% of the income and wealth distribution in thebaseline year. Robust standard errors in parentheses. (∗ : p < 0.1,∗∗ : p < 0.05,∗∗∗ : p < 0.01).

43

Table A7: Long-term treatment effects on self-reporting of any foreign income(1) (2) (3) (4) (5) (6)

No letter -0.038∗∗∗ -0.033∗∗∗ -0.023∗∗ -0.020∗ -0.023∗∗ -0.020∗

(0.010) (0.0097) (0.012) (0.011) (0.012) (0.011)

Fairness 0.016 0.014(0.012) (0.012)

Societal Benefits 0.0037 -0.00032(0.0090) (0.0085)

Detection 0.076∗∗∗ 0.076∗∗∗ 0.076∗∗∗ 0.076∗∗∗

(0.013) (0.012) (0.013) (0.012)

Moral 0.0068 0.0033(0.0085) (0.0081)

2011 outcome 0.13∗∗∗ 0.13∗∗∗ 0.13∗∗∗

(0.0077) (0.0076) (0.0076)

ACA-report baseline 0.013 0.013 0.013(0.021) (0.021) (0.021)

ACA-report follow-up 0.044 0.043 0.043(0.031) (0.032) (0.032)

Female 0.023∗∗∗ 0.023∗∗∗ 0.023∗∗∗

(0.0068) (0.0068) (0.0068)

Age > 60 yrs 0.31∗∗∗ 0.31∗∗∗ 0.31∗∗∗

(0.0080) (0.0079) (0.0080)

Norwegian citizen -0.061∗∗∗ -0.061∗∗∗ -0.061∗∗∗

(0.0074) (0.0074) (0.0074)

High SES 0.017∗∗ 0.017∗∗ 0.017∗∗

(0.0075) (0.0075) (0.0075)

Constant 0.27∗∗∗ 0.057∗∗∗ 0.25∗∗∗ 0.044∗∗∗ 0.25∗∗∗ 0.044∗∗∗

(0.0038) (0.0074) (0.0069) (0.0092) (0.0069) (0.0092)

F-test p on Moral treatments being equal: 0.27 0.18Observations 15708 15708 15708 15708 15708 15708R2 0.001 0.103 0.004 0.106 0.004 0.106

The table reports regressions based on equation 3, where the dependent variable is an indicatorvariable taking the value one if the taxpayer self-reports any foreign income reported in 2013. Es-timated effects are relative to the Base treatment. The indicator variables “No letter”, “Fairness”,“Societal Benefits”, and “Detection” take the value one if the taxpayer is in the respective treat-ment. The indicator variable “Moral” is one if the taxpayer is in the Fairness treatment or theSocial Benefits treatment. The reported F-test p is for the hypothesis that all Moral treatmentshave the same effect. Columns (2), (4), and (6) include the following controls: an indicator vari-able taking the value one if the taxpayer self-reported any foreign income for the baseline year, theamount of foreign income recorded in the ACA-reports for 2011 and 2012 (scaled in units of 1 000000 NOK), gender, age (an indicator variable taking the value one if the taxpayer is more than 60years), an indicator variable taking the value one if the tax payer is a Norwegian citizen, and anindicator variable of socio-economic status taking the value one if the taxpayer is in the upper 25%of the income and wealth distribution in the baseline year. Robust standard errors in parentheses.(∗ : p < 0.1,∗∗ : p < 0.05,∗∗∗ : p < 0.01).

44

Table A8: Long-term treatment effects on self-reported foreign income: Varioussubsamples

Level reported (in NOK) Reporting positive amount

Estimating sample: All A B A+B All A B A+B(1) (2) (3) (4) (5) (6) (7) (8)

No letter -1981.7 -405.7 3568.7 3380.5 -0.020∗ 0.0013 0.011 0.030∗

(3588.8) (4255.3) (4715.9) (4896.1) (0.011) (0.015) (0.013) (0.017)

Detection -872.1 335.4 -864.8 320.9 0.076∗∗∗ 0.054∗∗∗ 0.076∗∗∗ 0.054∗∗∗

(3023.9) (4212.8) (3024.7) (4214.6) (0.012) (0.015) (0.012) (0.015)

Moral 3651.4 7736.3 3646.1 7725.0 0.0033 0.0053 0.0033 0.0052(6015.4) (9734.7) (6016.5) (9739.1) (0.0081) (0.011) (0.0081) (0.011)

Baseline value 0.11∗∗ 0.072∗ 0.11∗∗ 0.070∗ 0.13∗∗∗ 0.10∗∗∗ 0.13∗∗∗ 0.100∗∗∗

(0.046) (0.040) (0.046) (0.040) (0.0076) (0.0095) (0.0078) (0.0097)

ACA report 2011 0.0075 0.084∗∗ 0.0073 0.083∗∗ 0.013 0.16∗∗∗ 0.012 0.15∗∗∗

(0.0099) (0.037) (0.0098) (0.037) (0.021) (0.037) (0.021) (0.036)

ACA report 2012 0.016 0.012 0.016 0.012 0.043 0.032 0.043 0.031(0.010) (0.010) (0.010) (0.0098) (0.032) (0.031) (0.032) (0.030)

Female -7534.1 -7505.2 -7776.1 -7740.0 0.023∗∗∗ 0.028∗∗∗ 0.023∗∗∗ 0.028∗∗∗

(4726.0) (6903.4) (4915.3) (7123.3) (0.0068) (0.0088) (0.0070) (0.0089)

Age > 60 yrs 20492.5∗∗∗ 29164.4∗∗∗ 20951.2∗∗∗ 29740.1∗∗∗ 0.31∗∗∗ 0.37∗∗∗ 0.31∗∗∗ 0.37∗∗∗

(3822.0) (5879.1) (3970.3) (6094.8) (0.0080) (0.010) (0.0081) (0.010)

Norwegian citizen -6188.8∗ 2537.8 -6210.8 2735.2 -0.061∗∗∗ -0.050∗∗∗ -0.061∗∗∗ -0.051∗∗∗

(3661.9) (6781.3) (3820.3) (7008.8) (0.0074) (0.0096) (0.0076) (0.0098)

High SES 12151.4∗∗ 14571.0∗ 12558.2∗∗ 14929.1∗ 0.017∗∗ 0.020∗∗ 0.019∗∗ 0.022∗∗

(5003.4) (8583.9) (5218.1) (8864.9) (0.0075) (0.0100) (0.0077) (0.010)

Constant 9954.2∗∗ 1148.7 9685.3∗∗ 867.6 0.044∗∗∗ 0.055∗∗∗ 0.044∗∗∗ 0.055∗∗∗

(4465.7) (8546.6) (4591.2) (8762.9) (0.0092) (0.012) (0.0093) (0.012)

Observations 15708 9688 15056 9371 15708 9688 15056 9371R2 0.006 0.006 0.006 0.005 0.106 0.144 0.107 0.147

The table reports regressions using the same specification as in Table A6 and Table A7, where Columns 1 and5 correspond to column 6 in Table A6 and Table A7. The remaining columns exclude participants with baselineforeign income recorded in ACA-reports in a range that might imply that they were targeted for differentiatedfollow (A) or participants that were part of the survey follow (B). Columns A+B exclude both groups. Robuststandard errors in parentheses (∗ : p < 0.1,∗∗ : p < 0.05,∗∗∗ : p < 0.01).

B Appendix for online publication only: Translationof all treatment letters

Figure B1: Base letter (I)

45

46

Figure B2: Base letter (II)

47

Figure B3: Fairness letter

48

Figure B4: Societal benefits letter (I)

49

Figure B5: Societal benefits letter (II)

This letter is the same as base letter (I), Figure B1, and it was combined with theattachment, Figure B6.

50

Din skatt finansierer viktige samfunnstjenester.

Figure B6: Attachment to Societal Benefits letters

The subtitle to the picture states that “Your taxes finance important public services.”Attachments were sent in Norwegian to all recipients of the relevant treatment let-ters.

51

Figure B7: Societal benefits letter (III)

This letter is the same as societal benefits (I) letter, Figure B4, and it was combinedwith the attachment, Figure B6.

52

Figure B8: Detection letter